Add Containerfile: build boco (musl) + slim kernel; runnable runner image

- 3-stage build: musl boco+boco-init (rust>=1.88, alpine:edge), linux-7.2
  bzImage from kernel/config (pinned tarball, sha256-verified), slim runtime
  with qemu/virtiofsd/podman.
- .containerignore: deny-all context (Cargo files, src/, init/, kernel/config)
  so the linux submodule no longer ships into every build.
- Verified end-to-end: kernel boots under KVM, ext4 rootfs, userspace init.
- NOTE: nested rootless podman cannot 'podman unshare -> container mount'
  (build path); boot path works as-is.

.containerignore

@@ -0,0 +1,10 @@
+# Build context: deny everything, then allow only what the Containerfile needs.
+# Keeps the linux submodule (and example/, .gitea/, README, etc.) out of the build.
+*
+!Cargo.toml
+!Cargo.lock
+!src/
+!src/**
+!init/
+!init/**
+!kernel/config

.gitmodules

@@ -0,0 +1,4 @@
+[submodule "linux"]
+	path = kernel/linux
+	url = https://github.com/torvalds/linux.git
+	shallow = true

Cargo.lock

@@ -66,15 +66,37 @@ checksum = "330a5ed07fa54e4702c9d6c4174f74427fc0ef6e214bbd677ae50a5099946470"
 
 [[package]]
 name = "base64"
-version = "0.22.1"
+version = "0.23.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "72b3254f16251a8381aa12e40e3c4d2f0199f8c6508fbecb9d91f575e0fbb8c6"
+checksum = "ac07cdecf99051d9a5238b80f35af32cdeba5b336e55d957b318b50137e18da5"
 
 [[package]]
 name = "bitflags"
-version = "2.13.1"
+version = "2.13.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b588b76d00fde79687d7646a9b5bdf3cc0f655e0bbd080335a95d7e96f3587da"
+checksum = "3ded4057c258ba199e2d26386d3af3780957ecaee6c4ef4041c6b4b8b97c0b06"
+
+[[package]]
+name = "boco"
+version = "0.1.0"
+dependencies = [
+ "anyhow",
+ "base64",
+ "clap",
+ "flate2",
+ "serde",
+ "serde_json",
+ "tempfile",
+ "xdg",
+]
+
+[[package]]
+name = "boco-init"
+version = "0.1.0"
+dependencies = [
+ "base64",
+ "nix",
+]
 
 [[package]]
 name = "cfg-if"
@@ -82,6 +104,12 @@ version = "1.0.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9330f8b2ff13f34540b44e946ef35111825727b38d33286ef986142615121801"
 
+[[package]]
+name = "cfg_aliases"
+version = "0.2.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f079e83a288787bcd14a6aea84cee5c87a67c5a3e660c30f557a3d24761b3527"
+
 [[package]]
 name = "clap"
 version = "4.6.6"
@@ -155,12 +183,13 @@ checksum = "da7c62ceae207dd37ea5b845da6a0696c799f85e97da1ab5b7910be3c1c80223"
 
 [[package]]
 name = "flate2"
-version = "1.1.9"
+version = "1.1.10"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "843fba2746e448b37e26a819579957415c8cef339bf08564fe8b7ddbd959573c"
+checksum = "6e634e2e0ebac1ee034020da1ca582e17ffe4e0f5e985823721e168928136dcb"
 dependencies = [
  "crc32fast",
  "miniz_oxide",
+ "zlib-rs",
 ]
 
 [[package]]
@@ -212,14 +241,26 @@ checksum = "cf8baf1c55e62ffcace7a9f06f4bd9cd3f0c4beb022d3b367256b91b87513d98"
 
 [[package]]
 name = "miniz_oxide"
-version = "0.8.9"
+version = "0.9.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "1fa76a2c86f704bdb222d66965fb3d63269ce38518b83cb0575fca855ebb6316"
+checksum = "b63fbc4a50860e98e7b2aa7804ded1db5cbc3aff9193adaff57a6931bf7c4b4c"
 dependencies = [
  "adler2",
  "simd-adler32",
 ]
 
+[[package]]
+name = "nix"
+version = "0.31.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cf20d2fde8ff38632c426f1165ed7436270b44f199fc55284c38276f9db47c3d"
+dependencies = [
+ "bitflags",
+ "cfg-if",
+ "cfg_aliases",
+ "libc",
+]
+
 [[package]]
 name = "once_cell"
 version = "1.21.4"
@@ -318,20 +359,6 @@ version = "0.3.10"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "3a219298ac11a56ea9a6d2120044824d6f01aeb034955e7af7bc16858527deea"
 
-[[package]]
-name = "slim"
-version = "0.1.0"
-dependencies = [
- "anyhow",
- "base64",
- "clap",
- "flate2",
- "serde",
- "serde_json",
- "tempfile",
- "xdg",
-]
-
 [[package]]
 name = "strsim"
 version = "0.11.1"
@@ -340,9 +367,9 @@ checksum = "7da8b5736845d9f2fcb837ea5d9e2628564b3b043a70948a3f0b778838c5fb4f"
 
 [[package]]
 name = "syn"
-version = "3.0.4"
+version = "3.0.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e6275cddf4610d1775e6d1fe9469b2e77d0f39fd98fb7450901b821e0c53649f"
+checksum = "12df2e0110f65b775f769bb17ef989067a1d931b2eb822bd4346631eeada89f9"
 dependencies = [
  "proc-macro2",
  "quote",
@@ -395,6 +422,12 @@ version = "3.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "2fb433233f2df9344722454bc7e96465c9d03bff9d77c248f9e7523fe79585b5"
 
+[[package]]
+name = "zlib-rs"
+version = "0.6.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "34b31d188d9d685a4f9c7b46d6e36631b07058d2cfe190267adce54dc230bf12"
+
 [[package]]
 name = "zmij"
 version = "1.0.23"

Cargo.toml

@@ -1,11 +1,19 @@
+[workspace]
+members = [".", "init"]
+
+[profile.release]
+opt-level = "z"
+lto = true
+strip = true
+
 [package]
-name = "slim"
+name = "boco"
 version = "0.1.0"
 edition = "2024"
 
 [dependencies]
 anyhow = "1.0.104"
-base64 = "0.22"
+base64 = "0.23"
 clap = { version = "4.6.6", features = ["derive"] }
 flate2 = "1.1.9"
 serde = { version = "1.0.229", features = ["derive"] }

Containerfile

@@ -0,0 +1,65 @@
+# boco runner: compiles boco (static musl) AND the shared boco kernel (vmlinuz)
+# with the slim config checked into the boco repo, then serves as the runtime
+# image for `boco build` / `boco boot` under podman.
+#
+# Runtime requirements (boco shells out to nested rootless podman):
+#   --device /dev/kvm      KVM accel for the VMs (entrypoint chmods it)
+#   --device /dev/net/tun  only if nested container runs need pasta networking
+#   volume at /home/boco/.local/share/boco   persistent registry (images+kernel)
+#   NO --privileged: the container's own userns (single mapping) is what makes
+#   nested `podman unshare` work; ignore_chown_errors handles unmapped uids.
+#
+# Kernel source: pinned tarball linux-7.2 (= 8d3ae59288f1e7d58d76558a6ee96d533bc5019f,
+# verified == refs/tags/v7.2{} via git ls-remote), sha256-checked.
+
+FROM alpine:edge AS build
+RUN apk add --no-cache rust cargo git
+WORKDIR /src
+COPY . .
+# alpine's rust targets x86_64-alpine-linux-musl natively (static, no glibc dep)
+RUN cargo build --release -p boco -p boco-init \
+ && strip target/release/boco target/release/boco-init
+
+FROM alpine:3.22 AS kernel
+# g++ for host tools, libelf for objtool, openssl for extract-cert
+RUN apk add --no-cache make gcc g++ musl-dev perl flex bison bc \
+      elfutils-dev openssl-dev findutils tar xz zstd curl linux-headers diffutils
+WORKDIR /kbuild
+# Pinned upstream tarball: linux-7.2.tar.xz, sha256
+# f9fef3d14c0df53819026f4be74459835c2a0b0dcbf5b5bbd9ea19f0829402b3.
+# (Equivalent to the upstream shallow pin 8d3ae59288f1e7d58d76558a6ee96d533bc5019f
+# = refs/tags/v7.2{} — verified via `git ls-remote`; the tarball avoids
+# a multi-GB shallow clone through flaky slirp NAT.)
+RUN curl -fsSLo /tmp/linux-7.2.tar.xz https://cdn.kernel.org/pub/linux/kernel/v7.x/linux-7.2.tar.xz \
+ && echo "f9fef3d14c0df53819026f4be74459835c2a0b0dcbf5b5bbd9ea19f0829402b3  /tmp/linux-7.2.tar.xz" | sha256sum -c - \
+ && tar -xJf /tmp/linux-7.2.tar.xz \
+ && mv linux-7.2 linux
+# Install the slim kernel config checked into the boco repo; olddefconfig
+# normalizes it against this tree and fails nothing silently afterwards.
+COPY kernel/config /kbuild/linux/.config
+RUN make -C linux olddefconfig HOSTCC=gcc CC=gcc \
+ && grep -qE '^(# )?CONFIG_MODULES' linux/.config
+RUN make -C linux -j"$(nproc)" bzImage HOSTCC=gcc CC=gcc \
+ && cp linux/arch/x86/boot/bzImage /vmlinuz \
+ && ls -lh /vmlinuz
+
+FROM alpine:3.22
+RUN apk add --no-cache \
+      qemu-system-x86_64 qemu-img virtiofsd \
+      podman fuse3 openssh-client ca-certificates iptables ip6tables \
+      e2fsprogs cpio passt \
+ && adduser -D -s /bin/sh boco \
+ && echo boco:100000:65536 > /etc/subuid \
+ && echo boco:100000:65536 > /etc/subgid \
+ && mkdir -p /etc/containers \
+ && printf '[storage]\ndriver = "vfs"\nrunroot = "/run/containers/storage"\ngraphroot = "/var/lib/containers/storage"\n[storage.options]\nignore_chown_errors = "true"\n' > /etc/containers/storage.conf \
+ && printf '#!/bin/sh\n[ -e /dev/kvm ] && chmod 666 /dev/kvm\nexec /usr/local/bin/boco "$@"\n' > /usr/local/bin/entrypoint.sh \
+ && chmod 755 /usr/local/bin/entrypoint.sh
+COPY --from=build /src/target/release/boco /usr/local/bin/boco
+COPY --from=build /src/target/release/boco-init /home/boco/.local/share/boco/boco-init
+COPY --from=kernel /vmlinuz /home/boco/.local/share/boco/vmlinuz
+RUN chmod 755 /usr/local/bin/boco /home/boco/.local/share/boco/boco-init /home/boco/.local/share/boco/vmlinuz
+ENV XDG_DATA_HOME=/home/boco/.local/share
+WORKDIR /data
+ENTRYPOINT ["/usr/local/bin/entrypoint.sh"]
+CMD ["image", "list"]

README.md

@@ -1,32 +1,41 @@
-# slim
+# boco
 
-Turning containers into bootable VMs
+Turn container images into bootable VMs
 
 ## Example
 
+```Dockerfile
+# ./example/alpine/Containerfile
+FROM alpine:latest
+CMD ["/bin/sh"]
+```
+
 ```sh
 # Build the container
-podman build ./example/alpine -t slim-alpine
+podman build ./example/alpine -t boco-alpine
 
-# Make the container bootable. slim injects /slim/init and /slim/exec,
+# Make the container bootable. boco injects a custom init program,
 # then packs the rootfs as a qcow2 disk.
-slim build qcow2 slim-alpine
+boco build qcow2 boco-alpine
 
-# Boot the image using QEMU. The container's CMD runs automatically.
+# Boot the image using QEMU, and drop into /bin/sh.
-slim run slim-alpine
+boco boot boco-alpine
-
-# Override the command at runtime (base64-encoded on the kernel cmdline):
-slim run slim-alpine --cmd '/bin/sh'
-
-# Override the command at build time (baked into /slim/exec):
-slim build qcow2 slim-alpine --cmd '/bin/sh'
 
 # Cleanup
-slim image rm slim-alpine
+boco image rm boco-alpine
-podman image rm slim-alpine
+podman image rm boco-alpine
+```
+
+Can also be used to boot a more traditional systemd-based VM:
+
+```Dockerfile
+FROM archlinux:latest
+
+# This works! Although you should probably add a user or some services.
+CMD ["/sbin/init"]
 ```
 
 ## Kernel
 
-slim boots VMs using a shared kernel at `$XDG_DATA_HOME/slim-rs/registry/vmlinuz`.
+boco boots VMs using a shared kernel at `$XDG_DATA_HOME/boco/registry/vmlinuz`.
-Place a suitable x86 bzImage there before running `slim run`.
+Place a suitable x86 bzImage there before running `boco boot`.

example/test.sh

@@ -1,12 +1,12 @@
 #!/bin/bash
-# Test script for slim - boots containers and verifies CMD inference/override.
+# Test script for boco - boots containers and verifies CMD inference/override.
 # Requires: podman, qemu-system-x86_64, KVM, curl, and a kernel at
-# $XDG_DATA_HOME/slim-rs/registry/vmlinuz.
+# $XDG_DATA_HOME/boco/registry/vmlinuz.
 set -u
 
 SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
 TARGET_DIR="${CARGO_TARGET_DIR:-$SCRIPT_DIR/../target}"
-SLIM_BIN="$TARGET_DIR/debug/slim"
+BOCO_BIN="$TARGET_DIR/debug/boco"
 TIMEOUT=120
 
 pass=0
@@ -34,7 +34,7 @@ check_output() {
 }
 
 cleanup() {
-    "$SLIM_BIN" image rm "$1" >/dev/null 2>&1 || true
+    "$BOCO_BIN" image rm "$1" >/dev/null 2>&1 || true
     podman image rm "$2" >/dev/null 2>&1 || true
 }
 
@@ -54,10 +54,10 @@ FROM $from
 $extra_setup
 CMD ["/bin/sh", "-c", "echo CMD_INFERRED_OK; wget -q -O /dev/null http://1.1.1.1 2>/dev/null && echo CONN_OK; wget -q -O /dev/null http://example.org 2>/dev/null && echo DNS_OK; poweroff -f"]
 EOF
-    img="slim-test-$distro-infer"
+    img="boco-test-$distro-infer"
     if podman build -t "$img" -f "$work/Containerfile.infer" >/dev/null 2>&1; then
-        "$SLIM_BIN" build qcow2 "$img" >/dev/null 2>&1
+        "$BOCO_BIN" build qcow2 "$img" >/dev/null 2>&1
-        output=$(timeout "$TIMEOUT" "$SLIM_BIN" run "$img" 2>&1 || true)
+        output=$(timeout "$TIMEOUT" "$BOCO_BIN" boot "$img" 2>&1 || true)
         check_output "$output" "CMD_INFERRED_OK" "$distro CMD-inferred"
         check_output "$output" "CONN_OK" "$distro connect to 1.1.1.1"
         check_output "$output" "DNS_OK" "$distro DNS lookup"
@@ -66,34 +66,34 @@ EOF
     fi
     cleanup "$img" "$img"
 
-    # === Test 2: slim run --cmd override ===
+    # === Test 2: boco boot --cmd override ===
-    echo "-- Test 2: slim run --cmd override"
+    echo "-- Test 2: boco boot --cmd override"
     cat > "$work/Containerfile.run" <<EOF
 FROM $from
 $extra_setup
 CMD ["/bin/sh", "-c", "echo SHOULD_NOT_APPEAR; poweroff -f"]
 EOF
-    img="slim-test-$distro-run"
+    img="boco-test-$distro-run"
     if podman build -t "$img" -f "$work/Containerfile.run" >/dev/null 2>&1; then
-        "$SLIM_BIN" build qcow2 "$img" >/dev/null 2>&1
+        "$BOCO_BIN" build qcow2 "$img" >/dev/null 2>&1
-        output=$(timeout "$TIMEOUT" "$SLIM_BIN" run "$img" --cmd 'echo RUN_OVERRIDE_OK; poweroff -f' 2>&1 || true)
+        output=$(timeout "$TIMEOUT" "$BOCO_BIN" boot "$img" --cmd 'echo RUN_OVERRIDE_OK; poweroff -f' 2>&1 || true)
-        check_output "$output" "RUN_OVERRIDE_OK" "$distro run --cmd override"
+        check_output "$output" "RUN_OVERRIDE_OK" "$distro boot --cmd override"
     else
-        report fail "$distro run --cmd override (podman build failed)"
+        report fail "$distro boot --cmd override (podman build failed)"
     fi
     cleanup "$img" "$img"
 
-    # === Test 3: slim build --cmd override ===
+    # === Test 3: boco build --cmd override ===
-    echo "-- Test 3: slim build --cmd override"
+    echo "-- Test 3: boco build --cmd override"
     cat > "$work/Containerfile.build" <<EOF
 FROM $from
 $extra_setup
 CMD ["/bin/sh", "-c", "echo SHOULD_NOT_APPEAR; poweroff -f"]
 EOF
-    img="slim-test-$distro-build"
+    img="boco-test-$distro-build"
     if podman build -t "$img" -f "$work/Containerfile.build" >/dev/null 2>&1; then
-        "$SLIM_BIN" build qcow2 "$img" --cmd 'echo BUILD_OVERRIDE_OK; poweroff -f' >/dev/null 2>&1
+        "$BOCO_BIN" build qcow2 "$img" --cmd 'echo BUILD_OVERRIDE_OK; poweroff -f' >/dev/null 2>&1
-        output=$(timeout "$TIMEOUT" "$SLIM_BIN" run "$img" 2>&1 || true)
+        output=$(timeout "$TIMEOUT" "$BOCO_BIN" boot "$img" 2>&1 || true)
         check_output "$output" "BUILD_OVERRIDE_OK" "$distro build --cmd override"
     else
         report fail "$distro build --cmd override (podman build failed)"
@@ -107,7 +107,7 @@ test_user() {
     echo "=== Testing USER directive ==="
 
     work="$(mktemp -d)"
-    img="slim-test-user"
+    img="boco-test-user"
 
     cat > "$work/Containerfile" <<'EOF'
 FROM alpine:latest
@@ -124,25 +124,25 @@ EOF
     fi
 
     echo "-- Test 1: build-time CMD runs as USER"
-    "$SLIM_BIN" build qcow2 "$img" >/dev/null 2>&1
+    "$BOCO_BIN" build qcow2 "$img" >/dev/null 2>&1
-    output=$(timeout "$TIMEOUT" "$SLIM_BIN" run "$img" 2>&1 || true)
+    output=$(timeout "$TIMEOUT" "$BOCO_BIN" boot "$img" 2>&1 || true)
     check_output "$output" "USER_BUILD_OK:1500:testuser" "USER build-time CMD runs as testuser"
 
-    echo "-- Test 2: run --cmd override runs as USER"
+    echo "-- Test 2: boot --cmd override runs as USER"
-    output=$(timeout "$TIMEOUT" "$SLIM_BIN" run "$img" --cmd 'echo USER_RUN_OK:$(id -u):$(whoami); poweroff -f' 2>&1 || true)
+    output=$(timeout "$TIMEOUT" "$BOCO_BIN" boot "$img" --cmd 'echo USER_RUN_OK:$(id -u):$(whoami); poweroff -f' 2>&1 || true)
-    check_output "$output" "USER_RUN_OK:1500:testuser" "USER run --cmd override runs as testuser"
+    check_output "$output" "USER_RUN_OK:1500:testuser" "USER boot --cmd override runs as testuser"
 
     cleanup "$img" "$img"
     rm -rf "$work"
 }
 
 test_mount() {
-    echo "=== Testing --mount (9p shares) ==="
+    echo "=== Testing --mount (virtio-fs shares) ==="
 
     work="$(mktemp -d)"
 
     # --- Image 1: no WORKINGDIR, test host:guest and multi-mount ---
-    img="slim-test-mount"
+    img="boco-test-mount"
     share_dir="$work/share"
     share_dir2="$work/share2"
     guest_dir="/mnt/guest"
@@ -164,9 +164,9 @@ EOF
         return
     fi
 
-    echo "-- Building slim VM..."
+    echo "-- Building boco VM..."
-    if ! "$SLIM_BIN" build qcow2 "$img" >/dev/null 2>&1; then
+    if ! "$BOCO_BIN" build qcow2 "$img" >/dev/null 2>&1; then
-        report fail "--mount (slim build failed)"
+        report fail "--mount (boco build failed)"
         cleanup "$img" "$img"
         rm -rf "$work"
         return
@@ -176,15 +176,13 @@ EOF
     share_canon2=$(readlink -f "$share_dir2")
 
     echo "-- Test 1: --mount host:guest (absolute guest path, multi-mount)..."
-    output=$(timeout "$TIMEOUT" "$SLIM_BIN" run "$img" \
+    output=$(timeout "$TIMEOUT" "$BOCO_BIN" boot "$img" \
         --mount "${share_dir}:${guest_dir}" \
         --mount "${share_dir2}:${guest_dir2}" \
-        --cmd "for f in /sys/bus/virtio/drivers/9pnet_virtio/virtio*/mount_tag; do [ -f \"\$f\" ] && echo \"\$f: \$(tr -d '\\0' < \"\$f\")\"; done; mount -v; cat ${guest_dir}/testfile.txt 2>/dev/null || echo NO_FILE; cat ${guest_dir2}/testfile2.txt 2>/dev/null || echo NO_FILE2; echo MOUNT_VERIFY_DONE; poweroff -f" \
+        --cmd "mount -v; cat ${guest_dir}/testfile.txt 2>/dev/null || echo NO_FILE; cat ${guest_dir2}/testfile2.txt 2>/dev/null || echo NO_FILE2; echo MOUNT_VERIFY_DONE; poweroff -f" \
         2>&1 || true)
 
-    check_output "$output" "mount_tag: slim0" "--mount 9p share detected"
+    check_output "$output" "type virtiofs" "--mount virtiofs filesystem in mount list"
-    check_output "$output" "mount_tag: slim1" "--mount second 9p share detected"
-    check_output "$output" "type 9p" "--mount 9p filesystem in mount list"
     check_output "$output" "hello from host" "--mount file accessible at guest path"
     check_output "$output" "second share" "--mount second file accessible at guest path"
     check_output "$output" "MOUNT_VERIFY_DONE" "--mount VM ran to completion"
@@ -192,7 +190,7 @@ EOF
     cleanup "$img" "$img"
 
     # --- Image 2: with WORKINGDIR, test relative guest path ---
-    img="slim-test-mount-wd"
+    img="boco-test-mount-wd"
     share_dir3="$work/share3"
 
     mkdir -p "$share_dir3"
@@ -211,32 +209,35 @@ EOF
         return
     fi
 
-    echo "-- Building slim VM..."
+    echo "-- Building boco VM..."
-    if ! "$SLIM_BIN" build qcow2 "$img" >/dev/null 2>&1; then
+    if ! "$BOCO_BIN" build qcow2 "$img" >/dev/null 2>&1; then
-        report fail "--mount relative (slim build failed)"
+        report fail "--mount relative (boco build failed)"
         cleanup "$img" "$img"
         rm -rf "$work"
         return
     fi
 
     echo "-- Test 2: --mount host:relative-guest (resolved against WORKINGDIR)..."
-    output=$(timeout "$TIMEOUT" "$SLIM_BIN" run "$img" \
+    output=$(timeout "$TIMEOUT" "$BOCO_BIN" boot "$img" \
         --mount "${share_dir3}:data" \
         --cmd "mount -v; cat /app/data/relfile.txt 2>/dev/null || echo NO_REL_FILE; echo REL_VERIFY_DONE; poweroff -f" \
         2>&1 || true)
 
-    check_output "$output" "type 9p" "--mount relative 9p filesystem in mount list"
+    check_output "$output" "type virtiofs" "--mount relative virtiofs filesystem in mount list"
     check_output "$output" "relative share" "--mount relative file accessible at WORKINGDIR/data"
     check_output "$output" "REL_VERIFY_DONE" "--mount relative VM ran to completion"
 
     cleanup "$img" "$img"
+    # Safety net: virtiofsd should exit when QEMU disconnects, but
+    # ensure no daemons linger if a boot failed mid-way.
+    pkill -f virtiofsd 2>/dev/null || true
     rm -rf "$work"
 }
 
 test_service() {
     echo "=== Testing nextcloud service ==="
 
-    img="slim-test-nextcloud"
+    img="boco-test-nextcloud"
     host_port=18080
     work="$(mktemp -d)"
 
@@ -252,16 +253,16 @@ EOF
         return
     fi
 
-    echo "-- Building slim VM..."
+    echo "-- Building boco VM..."
-    if ! "$SLIM_BIN" build qcow2 "$img" >/dev/null 2>&1; then
+    if ! "$BOCO_BIN" build qcow2 "$img" >/dev/null 2>&1; then
-        report fail "nextcloud service (slim build failed)"
+        report fail "nextcloud service (boco build failed)"
         cleanup "$img" "$img"
         rm -rf "$work"
         return
     fi
 
     echo "-- Booting VM with port forward (host :${host_port} -> guest :80)..."
-    setsid timeout 300 "$SLIM_BIN" run "$img" \
+    setsid timeout 300 "$BOCO_BIN" boot "$img" \
         --forward "tcp:0.0.0.0:${host_port}-:80" \
         --memory 2048M >/dev/null 2>&1 &
     vm_pid=$!
@@ -295,8 +296,63 @@ EOF
     rm -rf "$work"
 }
 
-echo "Building slim..."
+test_systemd() {
+    echo "=== Testing systemd as init ==="
+
+    work="$(mktemp -d)"
+    img="boco-test-systemd"
+
+    cat > "$work/boco-test.service" <<'SVC'
+[Unit]
+Description=boco systemd test
+After=multi-user.target
+
+[Service]
+Type=oneshot
+ExecStart=/bin/sh -c "echo SYSTEMD_TEST_OK > /dev/console; sleep 1; systemctl poweroff"
+
+[Install]
+WantedBy=multi-user.target
+SVC
+
+    cat > "$work/Containerfile" <<EOF
+FROM archlinux:latest
+COPY boco-test.service /etc/systemd/system/boco-test.service
+RUN mkdir -p /etc/systemd/system/multi-user.target.wants && \
+    ln -sf /etc/systemd/system/boco-test.service \
+       /etc/systemd/system/multi-user.target.wants/boco-test.service
+CMD ["/sbin/init"]
+EOF
+
+    echo "-- Building container image with systemd..."
+    if ! podman build -t "$img" "$work" >/dev/null 2>&1; then
+        report fail "systemd as init (podman build failed)"
+        rm -rf "$work"
+        return
+    fi
+
+    echo "-- Building boco VM..."
+    if ! "$BOCO_BIN" build qcow2 "$img" >/dev/null 2>&1; then
+        report fail "systemd as init (boco build failed)"
+        cleanup "$img" "$img"
+        rm -rf "$work"
+        return
+    fi
+
+    echo "-- Booting VM (systemd as PID 1)..."
+    output=$(timeout "$TIMEOUT" "$BOCO_BIN" boot "$img" 2>&1 || true)
+    check_output "$output" "SYSTEMD_TEST_OK" "systemd boots and runs service"
+    check_output "$output" "reboot: Power down" "systemd shuts down cleanly"
+
+    cleanup "$img" "$img"
+    rm -rf "$work"
+}
+
+echo "Building boco..."
 cargo build 2>&1
+cargo build --release -p boco-init --target x86_64-unknown-linux-musl 2>&1
+cp "$TARGET_DIR/x86_64-unknown-linux-musl/release/boco-init" \
+   "${XDG_DATA_HOME:-$HOME/.local/share}/boco/boco-init"
 
 test_distro "alpine" "alpine:latest" ""
 test_distro "archlinux" "archlinux:latest" "RUN pacman -Sy --noconfirm iproute2 wget; pacman -Sc --noconfirm"
@@ -307,6 +363,8 @@ test_mount
 
 test_service
 
+test_systemd
+
 echo ""
 echo "=== Results: $pass passed, $fail failed ==="
 [ "$fail" -eq 0 ]

init/Cargo.toml

@@ -0,0 +1,8 @@
+[package]
+name = "boco-init"
+version = "0.1.0"
+edition = "2024"
+
+[dependencies]
+base64 = "0.23"
+nix = { version = "0.31", features = ["mount", "fs", "process", "user", "reboot"] }

init/src/main.rs

@@ -0,0 +1,423 @@
+//! boco universal init.
+//!
+//! Injected by `boco build` at /boco/init and invoked via init=/boco/init.
+//! Sets up devices, filesystems, networking, and virtio-fs shares, then
+//! execs the container's CMD/ENTRYPOINT (or a runtime override) with
+//! execvp so the command becomes PID 1 directly.
+
+use base64::Engine;
+use nix::mount::{MsFlags, mount};
+use nix::sys::reboot::{RebootMode, reboot};
+use nix::sys::stat::{Mode, SFlag, makedev, mknod};
+use nix::sys::wait::waitpid;
+use nix::unistd::{ForkResult, Gid, Uid, User, execvp, fork, setgid, setuid};
+use std::env;
+use std::ffi::CString;
+use std::fs;
+use std::os::unix::fs::PermissionsExt;
+use std::path::Path;
+use std::process::Command;
+
+fn warn(msg: &str) {
+    eprintln!("boco-init: {msg}");
+}
+
+/// Run a fallible operation, logging on error.
+fn try_io<F, T>(label: &str, f: F)
+where
+    F: FnOnce() -> std::io::Result<T>,
+{
+    if let Err(e) = f() {
+        warn(&format!("{label}: {e}"));
+    }
+}
+
+/// Run a command, logging on error or non-zero exit.
+fn try_cmd(label: &str, cmd: &mut Command) {
+    match cmd.status() {
+        Ok(status) => {
+            if !status.success() {
+                warn(&format!("{label}: exited with {status}"));
+            }
+        }
+        Err(e) => warn(&format!("{label}: {e}")),
+    }
+}
+
+/// The kernel passes a minimal environment to init. Set a default PATH
+/// so that `ip` and other tools can be found by name.
+fn ensure_path() {
+    if env::var("PATH").is_err() {
+        // SAFETY: we are single-threaded before fork/exec.
+        unsafe {
+            env::set_var(
+                "PATH",
+                "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin",
+            );
+        }
+    }
+}
+
+fn find_in_path(prog: &str) -> Option<String> {
+    let path = env::var("PATH").ok()?;
+    for dir in path.split(':') {
+        let candidate = format!("{dir}/{prog}");
+        if Path::new(&candidate).exists() {
+            return Some(candidate);
+        }
+    }
+    None
+}
+
+fn try_mount(
+    source: Option<&str>,
+    target: &str,
+    fstype: Option<&str>,
+    flags: MsFlags,
+    data: Option<&str>,
+) {
+    if let Err(e) = mount(source, target, fstype, flags, data) {
+        warn(&format!("mount {target}: {e}"));
+    }
+}
+
+fn setup_filesystems() {
+    if !Path::new("/dev/console").exists() {
+        try_io("mknod /dev/console", || {
+            mknod(
+                "/dev/console",
+                SFlag::S_IFCHR,
+                Mode::S_IRUSR | Mode::S_IWUSR,
+                makedev(5, 1),
+            )
+            .map_err(std::io::Error::from)
+        });
+    }
+
+    let dirs = [
+        "/proc",
+        "/sys",
+        "/dev/pts",
+        "/dev/shm",
+        "/run",
+        "/tmp",
+        "/sys/fs/cgroup",
+    ];
+    for d in &dirs {
+        try_io(&format!("mkdir {d}"), || fs::create_dir_all(d));
+    }
+
+    try_mount(Some("proc"), "/proc", Some("proc"), MsFlags::empty(), None);
+    try_mount(Some("sysfs"), "/sys", Some("sysfs"), MsFlags::empty(), None);
+    try_mount(
+        Some("devpts"),
+        "/dev/pts",
+        Some("devpts"),
+        MsFlags::empty(),
+        None,
+    );
+    try_mount(
+        Some("tmpfs"),
+        "/dev/shm",
+        Some("tmpfs"),
+        MsFlags::empty(),
+        None,
+    );
+    try_mount(
+        Some("tmpfs"),
+        "/run",
+        Some("tmpfs"),
+        MsFlags::empty(),
+        Some("mode=755"),
+    );
+    try_mount(
+        Some("tmpfs"),
+        "/tmp",
+        Some("tmpfs"),
+        MsFlags::empty(),
+        Some("mode=1777"),
+    );
+
+    // cgroup2
+    try_mount(
+        Some("cgroup2"),
+        "/sys/fs/cgroup",
+        Some("cgroup2"),
+        MsFlags::empty(),
+        None,
+    );
+    if let Ok(controllers) = fs::read_to_string("/sys/fs/cgroup/cgroup.controllers") {
+        for c in controllers.split_whitespace() {
+            try_io("write cgroup.subtree_control", || {
+                fs::write("/sys/fs/cgroup/cgroup.subtree_control", format!("+{c}"))
+            });
+        }
+    }
+}
+
+fn setup_rootless_prereqs() {
+    // make-rshared /
+    if let Err(e) =
+        mount::<str, str, str, str>(None, "/", None, MsFlags::MS_REC | MsFlags::MS_SHARED, None)
+    {
+        warn(&format!("make-rshared /: {e}"));
+    }
+
+    // chmod u+s newuidmap newgidmap
+    for bin in &["/usr/bin/newuidmap", "/usr/bin/newgidmap"] {
+        match fs::metadata(bin) {
+            Ok(meta) => {
+                let mut perms = meta.permissions();
+                perms.set_mode(perms.mode() | 0o4000);
+                try_io(&format!("chmod u+s {bin}"), || {
+                    fs::set_permissions(bin, perms)
+                });
+            }
+            Err(e) if e.kind() == std::io::ErrorKind::NotFound => {}
+            Err(e) => warn(&format!("stat {bin}: {e}")),
+        }
+    }
+}
+
+fn setup_timezone() {
+    try_io("symlink timezone", || {
+        std::os::unix::fs::symlink("/usr/share/zoneinfo/Europe/Stockholm", "/etc/localtime")
+    });
+}
+
+fn setup_networking() {
+    let ip = find_in_path("ip").unwrap_or_else(|| "/sbin/ip".to_string());
+    try_cmd(
+        "ip link set lo up",
+        Command::new(&ip).args(["link", "set", "lo", "up"]),
+    );
+    try_cmd(
+        "ip link set eth0 up",
+        Command::new(&ip).args(["link", "set", "eth0", "up"]),
+    );
+    try_cmd(
+        "ip addr add 10.0.2.15/24 dev eth0",
+        Command::new(&ip).args(["addr", "add", "10.0.2.15/24", "dev", "eth0"]),
+    );
+    try_cmd(
+        "ip route add default via 10.0.2.2",
+        Command::new(&ip).args(["route", "add", "default", "via", "10.0.2.2"]),
+    );
+    try_io("write /etc/resolv.conf", || {
+        fs::write("/etc/resolv.conf", "nameserver 10.0.2.3\n")
+    });
+}
+
+fn setup_virtiofs_shares(workdir: &str) {
+    let cmdline = match fs::read_to_string("/proc/cmdline") {
+        Ok(c) => c,
+        Err(e) => {
+            warn(&format!("read /proc/cmdline: {e}"));
+            return;
+        }
+    };
+    let b64 = base64::engine::general_purpose::STANDARD;
+
+    for tok in cmdline.split_whitespace() {
+        if let Some(rest) = tok.strip_prefix("boco.mount=") {
+            let Some((tag, dest_b64)) = rest.split_once(':') else {
+                continue;
+            };
+            let dest = match b64.decode(dest_b64) {
+                Ok(bytes) => match String::from_utf8(bytes) {
+                    Ok(s) => s,
+                    Err(e) => {
+                        warn(&format!("virtiofs: invalid UTF-8 in guest path: {e}"));
+                        continue;
+                    }
+                },
+                Err(e) => {
+                    warn(&format!("virtiofs: base64 decode failed: {e}"));
+                    continue;
+                }
+            };
+
+            // Resolve relative guest paths against WORKINGDIR
+            let dest = if dest.starts_with('/') {
+                dest
+            } else {
+                format!("{}/{}", workdir, dest)
+            };
+
+            try_io(&format!("mkdir {dest}"), || fs::create_dir_all(&dest));
+            try_mount(Some(tag), &dest, Some("virtiofs"), MsFlags::empty(), None);
+        }
+    }
+}
+
+fn read_file_opt(path: &str) -> Option<String> {
+    fs::read_to_string(path)
+        .ok()
+        .filter(|s| !s.trim().is_empty())
+}
+
+fn split_null_delimited(data: &[u8]) -> Vec<String> {
+    data.split(|&b| b == 0)
+        .filter(|s| !s.is_empty())
+        .map(|s| String::from_utf8_lossy(s).into_owned())
+        .collect()
+}
+
+fn read_argv() -> Vec<String> {
+    // Check for runtime override: boco.cmd=<base64(null-delimited argv)>
+    let b64 = base64::engine::general_purpose::STANDARD;
+    if let Ok(cmdline) = fs::read_to_string("/proc/cmdline") {
+        for tok in cmdline.split_whitespace() {
+            if let Some(rest) = tok.strip_prefix("boco.cmd=")
+                && let Ok(decoded) = b64.decode(rest)
+            {
+                let argv = split_null_delimited(&decoded);
+                if !argv.is_empty() {
+                    return argv;
+                }
+            }
+        }
+    }
+
+    // Fall back to build-time argv from /boco/exec
+    match fs::read("/boco/exec") {
+        Ok(data) => {
+            let argv = split_null_delimited(&data);
+            if !argv.is_empty() {
+                return argv;
+            }
+        }
+        Err(e) => warn(&format!("read /boco/exec: {e}")),
+    }
+
+    warn("no command found, falling back to /bin/sh");
+    vec!["/bin/sh".to_string()]
+}
+
+fn apply_env() {
+    if let Ok(data) = fs::read("/boco/env") {
+        for entry in split_null_delimited(&data) {
+            if let Some((key, val)) = entry.split_once('=') {
+                // SAFETY: we are single-threaded before fork/exec.
+                unsafe { env::set_var(key, val) };
+            }
+        }
+    }
+}
+
+fn apply_workdir() {
+    if let Some(dir) = read_file_opt("/boco/workdir") {
+        try_io("set working directory", || env::set_current_dir(&dir));
+    }
+}
+
+struct UserSpec {
+    uid: Uid,
+    gid: Gid,
+}
+
+fn resolve_user(spec: &str) -> Option<UserSpec> {
+    let (user_part, gid_part) = match spec.split_once(':') {
+        Some((u, g)) => (u, Some(g)),
+        None => (spec, None),
+    };
+
+    let (uid, default_gid) = if let Ok(numeric) = user_part.parse::<u32>() {
+        let user = User::from_uid(Uid::from_raw(numeric)).ok().flatten();
+        let gid = user
+            .as_ref()
+            .map(|u| u.gid)
+            .unwrap_or(Gid::from_raw(numeric));
+        (Uid::from_raw(numeric), gid)
+    } else {
+        let user = User::from_name(user_part).ok().flatten()?;
+        (user.uid, user.gid)
+    };
+
+    let gid = match gid_part {
+        Some(g) => match g.parse::<u32>() {
+            Ok(n) => Gid::from_raw(n),
+            Err(_) => default_gid,
+        },
+        None => default_gid,
+    };
+
+    Some(UserSpec { uid, gid })
+}
+
+fn exec_command(argv: &[String], user: Option<UserSpec>) -> ! {
+    let cstrings: Vec<CString> = argv
+        .iter()
+        .map(|s| CString::new(s.as_str()).unwrap_or_default())
+        .collect();
+    let c_prog = cstrings[0].clone();
+    let c_refs: Vec<&CString> = cstrings.iter().collect();
+
+    match user {
+        None => {
+            // No privilege drop: exec directly, PID 1 becomes the command.
+            let Err(e) = execvp(&c_prog, &c_refs);
+            warn(&format!("execvp {}: {e}", argv[0]));
+            let _ = reboot(RebootMode::RB_POWER_OFF);
+            std::process::exit(1);
+        }
+        Some(user) => {
+            // Drop privileges in a child so PID 1 (root) can poweroff afterwards.
+            match unsafe { fork() } {
+                Ok(ForkResult::Child) => {
+                    // setgid before setuid to avoid losing privileges
+                    if let Err(e) = setgid(user.gid) {
+                        warn(&format!("setgid: {e}"));
+                    }
+                    if let Err(e) = setuid(user.uid) {
+                        warn(&format!("setuid: {e}"));
+                    }
+                    let Err(e) = execvp(&c_prog, &c_refs);
+                    warn(&format!("execvp {}: {e}", argv[0]));
+                    std::process::exit(127);
+                }
+                Ok(ForkResult::Parent { child }) => {
+                    let _ = waitpid(child, None);
+                    let _ = reboot(RebootMode::RB_POWER_OFF);
+                    std::process::exit(0);
+                }
+                Err(e) => {
+                    warn(&format!("fork: {e}"));
+                    let _ = reboot(RebootMode::RB_POWER_OFF);
+                    std::process::exit(1);
+                }
+            }
+        }
+    }
+}
+
+fn main() {
+    ensure_path();
+
+    // === Devices & special filesystems ===
+    setup_filesystems();
+
+    // === Rootless container prerequisites (best-effort) ===
+    setup_rootless_prereqs();
+
+    // === Timezone ===
+    setup_timezone();
+
+    // === Networking ===
+    setup_networking();
+
+    // === virtio-fs shares ===
+    let workdir = read_file_opt("/boco/workdir").unwrap_or_default();
+    setup_virtiofs_shares(&workdir);
+
+    // === Environment & working directory ===
+    apply_env();
+    apply_workdir();
+
+    // === Resolve user for privilege drop ===
+    let user = read_file_opt("/boco/user").and_then(|spec| resolve_user(&spec));
+
+    // === Execute the configured command ===
+    let argv = read_argv();
+    exec_command(&argv, user);
+}

src/build.rs

@@ -1,4 +1,4 @@
-//! Build: mount a podman image, inject /slim/ scripts, and pack the rootfs
+//! Build: mount a podman image, inject /boco/ scripts, and pack the rootfs
 //! as a gzipped cpio initrd or qcow2 disk in the registry.
 
 use anyhow::{Context, Result, bail};
@@ -41,7 +41,7 @@ pub(crate) fn build(
     }: BuildCmd,
 ) -> Result<()> {
     let image = &name;
-    let container = format!("slim-build-{}", image.replace(':', "-"));
+    let container = format!("boco-build-{}", image.replace(':', "-"));
 
     let mount_path = mount_container(image, &container)?;
     println!("Mounted at: {}", mount_path.display());
@@ -75,16 +75,18 @@ fn build_inner(
     cmd: Option<String>,
 ) -> Result<()> {
     let config = inject::inspect_config(image)?;
-    let command = cmd.unwrap_or_else(|| inject::infer_command(&config));
+    let argv = match cmd {
-    let exec_script =
+        Some(s) => vec!["/bin/sh".into(), "-c".into(), s],
-        inject::build_exec_script(&command, &config.env, config.working_dir.as_deref());
+        None => inject::infer_argv(&config),
+    };
     inject::inject(
         mount_path,
-        &exec_script,
+        &argv,
+        &config.env,
         config.user.as_deref(),
         config.working_dir.as_deref(),
     )?;
-    println!("Injected /slim/ (init + exec)");
+    println!("Injected /boco/ (init + exec)");
 
     match kind {
         ImageKind::Initrd => build_initrd(image, mount_path),

src/inject.rs

@@ -1,15 +1,14 @@
-//! Inject: inspect a podman image's config, infer the default command, and
+//! Inject: inspect a podman image's config, infer the default command argv,
-//! inject distro-agnostic /slim/ scripts (init + exec) into the mounted rootfs.
+//! and inject the boco init binary + null-delimited argv/env files into the
+//! mounted rootfs.
 
 use anyhow::{Context, Result, anyhow};
 use serde::Deserialize;
-use std::fmt::Write as _;
 use std::fs;
 use std::path::Path;
 
 use crate::command::cmd;
-
+use crate::registry::init_binary_path;
-const SLIM_INIT: &str = include_str!("scripts/slim-init.sh");
 
 #[derive(Debug, Default, Deserialize)]
 #[serde(rename_all = "PascalCase")]
@@ -40,85 +39,76 @@ pub fn inspect_config(image: &str) -> Result<Config> {
     Ok(config)
 }
 
-/// Infer the command string from the image config.
+/// Infer the command argv from the image config.
 ///
 /// Concatenates ENTRYPOINT + CMD (Docker semantics). If neither is present,
-/// falls back to `/bin/sh`.
+/// falls back to `["/bin/sh"]`.
-pub fn infer_command(config: &Config) -> String {
+pub fn infer_argv(config: &Config) -> Vec<String> {
-    let parts = config.entrypoint.iter().chain(config.cmd.iter());
+    let parts: Vec<String> = config
-    let parts: Vec<_> = parts.map(|s| s.as_str()).collect();
+        .entrypoint
-
-    match &parts[..] {
-        [] => "/bin/sh".into(),
-        ["/bin/sh" | "sh", "-c", cmd] => cmd.to_string(),
-        _ => shell_join(&parts),
-    }
-}
-
-fn shell_escape(s: &str) -> String {
-    format!("'{}'", s.replace('\'', "'\\''"))
-}
-
-fn shell_join(parts: &[&str]) -> String {
-    parts
         .iter()
-        .map(|p| shell_escape(p))
+        .chain(config.cmd.iter())
-        .collect::<Vec<_>>()
+        .cloned()
-        .join(" ")
+        .collect();
+
+    if parts.is_empty() {
+        vec!["/bin/sh".into()]
+    } else {
+        parts
+    }
 }
 
-/// Generate the /slim/exec script from a command string, env vars, and working dir.
+/// Join items as null-delimited bytes.
-pub fn build_exec_script(command: &str, env: &[String], working_dir: Option<&str>) -> String {
+pub fn null_delimited(items: &[String]) -> Vec<u8> {
-    let mut lines = String::from("#!/bin/sh\n");
+    let mut data = Vec::new();
-    if let Some(dir) = working_dir.filter(|d| !d.is_empty()) {
+    for item in items {
-        _ = writeln!(&mut lines, "cd {} 2>/dev/null", shell_escape(dir));
+        data.extend_from_slice(item.as_bytes());
+        data.push(0);
     }
-    for var in env {
+    data
-        if let Some((key, val)) = var.split_once('=') {
-            _ = writeln!(
-                &mut lines,
-                "export {}={}",
-                shell_escape(key),
-                shell_escape(val)
-            );
-        }
-    }
-    _ = writeln!(&mut lines, "exec /bin/sh -c {}", shell_escape(command));
-    lines
 }
 
-/// Install `content` into the mounted rootfs at `<mount>/slim/<name>` with
+/// Install `content` into the mounted rootfs at `<mount>/boco/<name>` with
 /// the given mode.
-fn install_into_rootfs(mount: &str, name: &str, mode: &str, content: &str) -> Result<()> {
+fn install_into_rootfs(mount: &str, name: &str, mode: &str, content: &[u8]) -> Result<()> {
     let temp = tempfile::NamedTempFile::new()?;
     fs::write(temp.path(), content)?;
     let src = temp.path().to_str().context("temp path is not UTF-8")?;
-    let dest = format!("{mount}/slim/{name}");
+    let dest = format!("{mount}/boco/{name}");
     cmd(&[
         "podman", "unshare", "--", "install", "-D", "-m", mode, src, &dest,
     ])?;
     Ok(())
 }
 
-/// Inject /slim/init, /slim/exec, and optionally /slim/user and /slim/workdir
+/// Inject /boco/init (binary), /boco/exec (null-delimited argv), /boco/env
-/// into a mounted container image rootfs.
+/// (null-delimited env), and optionally /boco/user and /boco/workdir into a
+/// mounted container image rootfs.
 pub fn inject(
     mount_path: &Path,
-    exec_script: &str,
+    argv: &[String],
+    env: &[String],
     user: Option<&str>,
     working_dir: Option<&str>,
 ) -> Result<()> {
     let mount_str = mount_path.to_str().context("mount path is not UTF-8")?;
 
-    install_into_rootfs(mount_str, "init", "755", SLIM_INIT)?;
+    let init_binary = fs::read(init_binary_path()?).context("Failed to read boco-init binary")?;
-    install_into_rootfs(mount_str, "exec", "755", exec_script)?;
+    let exec_data = null_delimited(argv);
+    let env_data = null_delimited(env);
+
+    install_into_rootfs(mount_str, "init", "755", &init_binary)?;
+    install_into_rootfs(mount_str, "exec", "644", &exec_data)?;
+    if !env_data.is_empty() {
+        install_into_rootfs(mount_str, "env", "644", &env_data)?;
+    }
 
     if let Some(user) = user.filter(|u| !u.is_empty()) {
-        install_into_rootfs(mount_str, "user", "644", user)?;
+        install_into_rootfs(mount_str, "user", "644", user.as_bytes())?;
     }
 
     if let Some(dir) = working_dir.filter(|d| !d.is_empty()) {
-        install_into_rootfs(mount_str, "workdir", "644", dir)?;
+        install_into_rootfs(mount_str, "workdir", "644", dir.as_bytes())?;
     }
 
     Ok(())

src/main.rs

@@ -5,11 +5,12 @@ mod image;
 mod inject;
 mod qemu;
 mod registry;
+mod virtiofs;
 
 use anyhow::Result;
 use clap::{Parser, Subcommand};
 
-use crate::{build::BuildCmd, image::ImageCmd, qemu::RunCmd};
+use crate::{build::BuildCmd, image::ImageCmd, qemu::BootCmd};
 
 #[derive(Parser, Debug)]
 #[command(about = "Turn containers into bootable VMs")]
@@ -23,8 +24,8 @@ enum Commands {
     /// Build initrd from an image (podman image must exist)
     Build(BuildCmd),
     /// Launch a previously built VM via QEMU.
-    Run(RunCmd),
+    Boot(BootCmd),
-    /// List slim images in the registry
+    /// List boco images in the registry
     #[command(subcommand)]
     Image(ImageCmd),
 }
@@ -35,8 +36,8 @@ fn main() -> Result<()> {
         Commands::Build(cmd) => {
             build::build(cmd)?;
         }
-        Commands::Run(cmd) => {
+        Commands::Boot(cmd) => {
-            qemu::run(cmd)?;
+            qemu::boot(cmd)?;
         }
         Commands::Image(cmd) => {
             image::run(cmd)?;

src/qemu.rs

@@ -1,9 +1,4 @@
-//! QEMU: boot a registry entry with qemu-system-x86_64 using direct kernel
+//! See [`boot`]
-//! boot (-kernel/-initrd or -drive). The kernel is shared across all VMs
-//! at $XDG_DATA_HOME/slim-rs/registry/vmlinuz. A universal /slim/init script (injected
-//! at build time) handles VM bootstrap; /slim/exec runs the container's
-//! CMD/ENTRYPOINT. Runtime overrides are delivered as `slim.cmd=<base64>`
-//! on the kernel cmdline.
 
 use anyhow::{Context, Result, bail};
 use base64::Engine;
@@ -12,11 +7,12 @@ use std::{fmt::Write as _, process::Command};
 
 use crate::{
     forward::PortForward,
-    registry::{registry_base_dir, registry_dir},
+    registry::{registry_dir, vmlinuz_path},
+    virtiofs::{self, VirtiofsShare},
 };
 
 #[derive(Args, Debug)]
-pub struct RunCmd {
+pub struct BootCmd {
     /// Image tag / registry subdir name
     name: String,
 
@@ -29,11 +25,11 @@ pub struct RunCmd {
     forward: Vec<PortForward>,
 
     /// Override the command to exec in the VM (base64-encoded on the kernel
-    /// cmdline as slim.cmd=<b64>). Overrides the CMD inferred at build time.
+    /// cmdline as `boco.cmd=<b64>`). Overrides the CMD inferred at build time.
     #[clap(long)]
     cmd: Option<String>,
 
-    /// Share a host directory into the VM via 9p. Format:
+    /// Share a host directory into the VM via virtio-fs. Format:
     /// `<host-path>` or `<host-path>:<guest-path>`. When the guest path is
     /// omitted, the host path is used. Non-absolute guest paths are
     /// relative to the image's WORKINGDIR. Can be repeated.
@@ -41,23 +37,27 @@ pub struct RunCmd {
     mount: Vec<String>,
 }
 
-pub(crate) fn run(
+/// Invoke `qemu` to boot a boco image. See [`BootCmd`] for options
-    RunCmd {
+///
+/// This uses qemu's direct-kernel-boot with `-kernel`, so it does not require a bootloader.
+/// The boco image is just an ext4 rootfs.
+pub(crate) fn boot(
+    BootCmd {
         name,
         memory,
         forward,
         cmd,
         mount,
-    }: RunCmd,
+    }: BootCmd,
 ) -> Result<()> {
     let reg_dir = registry_dir(&name)?;
-    let vmlinuz_path = registry_base_dir()?.join("vmlinuz");
+    let vmlinuz_path = vmlinuz_path()?;
     let initrd_path = reg_dir.join("initrd");
     let qcow2_path = reg_dir.join("image.qcow2");
     if !vmlinuz_path.exists() {
         // TODO: guide user in how to set up a kernel
         bail!(
-            "No kernel available. Place a vmlinuz at {}",
+            "No kernel available. Expected vmlinuz at {}",
             vmlinuz_path.display()
         );
     }
@@ -67,7 +67,7 @@ pub(crate) fn run(
         "rw".into(),
         "earlyprintk=serial".into(),
         "nokaslr".into(),
-        "init=/slim/init".into(),
+        "init=/boco/init".into(),
         "devtmpfs.mount=1".into(), // Automatically mount /dev at boot
     ];
 
@@ -81,43 +81,65 @@ pub(crate) fn run(
         fs_args = vec!["-initrd".to_string(), initrd_str];
         cmdline.push("root=/dev/ram0".into());
     } else {
-        bail!("Registry missing rootfs/initrd for '{name}'. Run `slim build` first.");
+        bail!("Registry missing rootfs/initrd for '{name}'. Run `boco build` first.");
     }
 
     if let Some(cmd) = &cmd {
-        let encoded = base64::engine::general_purpose::STANDARD.encode(cmd.as_bytes());
+        let argv = ["/bin/sh".to_string(), "-c".to_string(), cmd.clone()];
-        cmdline.push(format!("slim.cmd={encoded}"));
+        let data: Vec<u8> = argv.join("\0").into_bytes();
+        let encoded = base64::engine::general_purpose::STANDARD.encode(&data);
+        cmdline.push(format!("boco.cmd={encoded}"));
     }
 
-    // Build 9p shares for each --mount.  Tags are short (slim0, slim1, …)
+    // Build virtio-fs shares for each `--mount`.  Each share gets its own
-    // because 9p mount_tag has a ~31-byte limit.  The guest destination path
+    // virtiofsd daemon under `podman unshare` (rootless UID mapping) and a
-    // is passed on the kernel cmdline as slim.mount=<tag>:<base64(guest_path)>.
+    // matching `-chardev socket` + `-device vhost-user-fs-pci` pair on the
-    // If no guest path is specified, the host path is used as the guest path.
+    // QEMU side. The guest destination path is passed on the kernel cmdline
-    let mut virtfs_args: Vec<String> = Vec::new();
+    // as `boco.mount=<tag>:<base64(guest_path)>` and resolved by /boco/init.
+    let mut shares: Vec<VirtiofsShare> = Vec::with_capacity(mount.len());
+    let mut fsdev_args: Vec<String> = Vec::new();
     for (i, spec) in mount.iter().enumerate() {
         let (host_path, guest_path) = match spec.split_once(':') {
-            Some((h, g)) => (h, g),
+            Some((h, g)) => (h, g.to_string()),
-            None => (spec.as_str(), spec.as_str()),
+            None => (spec.as_str(), spec.to_string()),
         };
-        let canonical = std::path::Path::new(host_path)
+        let share = virtiofs::spawn_share(i, host_path, guest_path)?;
-            .canonicalize()
+        let tag = share.tag.clone();
-            .with_context(|| format!("Cannot resolve mount path '{host_path}'"))?;
+        let chardev = format!("char{i}");
-        let host_str = canonical
+        let socket = share
+            .socket_path()
             .to_str()
-            .context("Mount path is not valid UTF-8")?;
+            .context("socket path is not UTF-8")?
-        let tag = format!("slim{i}");
+            .to_string();
-        let dest_b64 = base64::engine::general_purpose::STANDARD.encode(guest_path.as_bytes());
+        let dest_b64 =
+            base64::engine::general_purpose::STANDARD.encode(share.guest_path.as_bytes());
 
-        // QEMU's QemuOpts splits on commas — escape literal commas in the
+        fsdev_args.push("-chardev".into());
-        // path as ",," per QEMU convention.
+        fsdev_args.push(format!("socket,id={chardev},path={socket}"));
-        let host_escaped = host_str.replace(',', ",,");
+        fsdev_args.push("-device".into());
-        virtfs_args.push("-virtfs".into());
+        fsdev_args.push(format!(
-        virtfs_args.push(format!(
+            "vhost-user-fs-pci,queue-size=1024,chardev={chardev},tag={tag}"
-            "local,path={host_escaped},mount_tag={tag},security_model=mapped-xattr"
         ));
-        cmdline.push(format!("slim.mount={tag}:{dest_b64}"));
+        cmdline.push(format!("boco.mount={tag}:{dest_b64}"));
+
+        shares.push(share);
     }
 
+    // vhost-user devices require shared guest RAM so virtiofsd (a separate
+    // process) can DMA into it.
+    let memory_args: Vec<String> = if shares.is_empty() {
+        vec!["-m".into(), memory.clone()]
+    } else {
+        vec![
+            "-m".into(),
+            memory.clone(),
+            "-object".into(),
+            format!("memory-backend-memfd,id=mem,size={memory},share=on"),
+            "-numa".into(),
+            "node,memdev=mem".into(),
+        ]
+    };
+
     println!("Booting {name} from registry: {}", reg_dir.display());
 
     let cmdline = cmdline.join(" ");
@@ -134,11 +156,11 @@ pub(crate) fn run(
     command
         .args(["-accel", "kvm"])
         .args(["-cpu", "host", "-smp", "cpus=8"])
-        .args(["-m", &memory])
+        .args(&memory_args)
         .arg("-kernel")
         .arg(vmlinuz_path.as_os_str())
         .args(&fs_args)
-        .args(&virtfs_args)
+        .args(&fsdev_args)
         .args(["-snapshot"])
         .args(["-no-reboot"])
         .args(["-append", &cmdline])
@@ -151,6 +173,9 @@ pub(crate) fn run(
         .status()
         .context("Failed to launch qemu-system-x86_64")?;
 
+    // Ensure virtiofsd daemons are torn down even on error returns below.
+    drop(shares);
+
     match status.code() {
         Some(0) => println!("QEMU exited cleanly."),
         Some(126) | Some(127) => {

src/registry.rs

@@ -1,12 +1,12 @@
 //! Registry: local storage of built VM artifacts under
-//! `$XDG_DATA_HOME/slim-rs/registry/<image>/`.
+//! `$XDG_DATA_HOME/boco/registry/<image>/`.
 
 use anyhow::{Context, Result, anyhow, bail};
 use std::{fs, io, path::PathBuf};
 
 /// Validate a user-supplied image name before it is used in a registry path.
 ///
-/// The name becomes a directory under `$XDG_DATA_HOME/slim-rs/registry/`, so
+/// The name becomes a directory under `$XDG_DATA_HOME/boco/registry/`, so
 /// path separators and `.`/`..` would allow escaping that directory.
 /// Allow-list: ASCII alphanumerics plus `.`, `_`, `:`, `-` (covers image
 /// refs like `alpine:3.15`).
@@ -26,12 +26,24 @@ pub(crate) fn validate_image_name(name: &str) -> Result<()> {
     Ok(())
 }
 
+pub(crate) fn vmlinuz_path() -> Result<PathBuf> {
+    let base = xdg::BaseDirectories::with_prefix("boco");
+    base.place_data_file("vmlinuz")
+        .context("Failed to create XDG_DATA_HOME subdirectory")
+}
+
 pub(crate) fn registry_base_dir() -> Result<PathBuf> {
-    let base = xdg::BaseDirectories::with_prefix("slim-rs");
+    let base = xdg::BaseDirectories::with_prefix("boco");
     base.create_data_directory("registry")
         .context("Failed to create XDG_DATA_HOME subdirectory")
 }
 
+pub(crate) fn init_binary_path() -> Result<PathBuf> {
+    let base = xdg::BaseDirectories::with_prefix("boco");
+    base.get_data_file("boco-init")
+        .context("boco-init not found in XDG data dir")
+}
+
 pub(crate) fn registry_dir(image: &str) -> Result<PathBuf> {
     validate_image_name(image)?;
     let base = registry_base_dir()?;
@@ -56,7 +68,7 @@ mod tests {
     fn accepts_valid_image_names() {
         for name in [
             "alpine:3.15",
-            "slim",
+            "boco",
             "mock-vm",
             "my_vm",
             "a.b.c",

src/scripts/slim-init.sh

@@ -1,121 +0,0 @@
-#!/bin/sh
-# slim universal init - distro-agnostic VM bootstrap.
-# Injected by `slim build` at /slim/init and invoked via init=/slim/init.
-
-# === Helpers ===
-b64dec() {
-    printf '%s' "$1" | base64 -d 2>/dev/null || printf '%s' "$1" | openssl base64 -d 2>/dev/null
-}
-
-# === Devices & special filesystems ===
-[ -c /dev/console ] || mknod -m 600 /dev/console c 5 1
-mkdir -p /proc /sys /dev/pts /dev/shm
-mount -t proc proc /proc
-mount -t sysfs sysfs /sys
-mount -t devpts devpts /dev/pts
-mount -t tmpfs tmpfs /dev/shm
-mount -t tmpfs tmpfs /run  -o mode=755
-mount -t tmpfs tmpfs /tmp  -o mode=1777
-
-# === cgroup2 ===
-mkdir -p /sys/fs/cgroup
-mount -t cgroup2 none /sys/fs/cgroup 2>/dev/null || echo "mount cgroup2 failed"
-# shellcheck disable=SC2013 # word-splitting is intentional: controllers are space-separated
-for c in $(cat /sys/fs/cgroup/cgroup.controllers 2>/dev/null); do
-    echo "+$c" > /sys/fs/cgroup/cgroup.subtree_control 2>/dev/null || echo "enable cgroup controller $c failed"
-done
-
-# === Rootless container prerequisites (best-effort) ===
-mount --make-rshared / 2>/dev/null || echo "make-rshared / failed"
-chmod u+s /usr/bin/newuidmap /usr/bin/newgidmap 2>/dev/null || echo "chmod newuidmap/newgidmap failed"
-
-# === Timezone ===
-ln -sf /usr/share/zoneinfo/Europe/Stockholm /etc/localtime 2>/dev/null || echo "set timezone failed"
-
-# === Networking (QEMU slirp: guest 10.0.2.15/24, gw 10.0.2.2, dns 10.0.2.3) ===
-ip link set lo up 2>/dev/null || echo "ip link set lo up failed"
-ip link set eth0 up 2>/dev/null || echo "ip link set eth0 up failed"
-ip addr add 10.0.2.15/24 dev eth0 2>/dev/null || echo "ip addr add eth0 failed"
-ip route add default via 10.0.2.2 2>/dev/null || echo "ip route add default failed"
-echo nameserver 10.0.2.3 > /etc/resolv.conf
-
-# === 9p shares (host directories shared via --mount) ===
-# Each share is passed as slim.mount=<tag>:<base64(guest_path)> on the
-# kernel cmdline.  The 9p tag is short (e.g. slim0) because mount_tag
-# has a ~31-byte limit; the guest path is base64-encoded.
-# Non-absolute guest paths are resolved against /slim/workdir (the
-# image's WORKINGDIR).
-SLIM_WORKDIR=""
-[ -f /slim/workdir ] && SLIM_WORKDIR=$(cat /slim/workdir 2>/dev/null)
-# shellcheck disable=SC2013 # word-splitting is intentional: cmdline tokens are space-separated
-for tok in $(cat /proc/cmdline 2>/dev/null); do
-    case "$tok" in
-        slim.mount=*)
-            v=${tok#slim.mount=}
-            tag=${v%%:*}
-            dest_b64=${v#*:}
-            dest=$(b64dec "$dest_b64") || echo "base64 decode failed"
-            if [ -n "$tag" ] && [ -n "$dest" ]; then
-                # Resolve relative paths against the image's WORKINGDIR
-                case "$dest" in
-                    /*) ;;
-                    *) dest="${SLIM_WORKDIR:-/}/${dest}" ;;
-                esac
-                mkdir -p "$dest"
-                mount -t 9p "$tag" "$dest" -o trans=virtio,version=9p2000.L 2>/dev/null \
-                    && echo "mounted 9p '$tag' -> '$dest'" \
-                    || echo "failed to mount 9p '$tag' -> '$dest'"
-            fi
-            ;;
-    esac
-done
-
-# === Drop privileges (respect USER directive from image config) ===
-# /slim/user may contain a username, uid, or uid:gid (OCI image spec).
-# Numeric uids are resolved to a username via /etc/passwd because BusyBox
-# su does not accept numeric arguments.
-SLIM_USER=""
-[ -f /slim/user ] && SLIM_USER=$(cat /slim/user 2>/dev/null)
-
-# Resolve a numeric uid to a username from /etc/passwd.
-slim_resolve_user() {
-    _u="$1"
-    case "$_u" in
-        *[!0-9]*) printf '%s' "$_u" ;;
-        *)
-            _resolved=$(awk -F: -v uid="$_u" '$3==uid{print $1; exit}' /etc/passwd 2>/dev/null)
-            [ -n "$_resolved" ] && printf '%s' "$_resolved" || printf '%s' "$_u"
-            ;;
-    esac
-}
-
-# Exec a command, dropping privileges if SLIM_USER is set.
-# When dropping privileges, run as a child (not exec) so PID 1 stays root
-# and can poweroff after the command exits — non-root cannot call poweroff.
-slim_exec() {
-    _cmd="$1"
-    if [ -z "$SLIM_USER" ]; then
-        exec /bin/sh -c "$_cmd"
-    fi
-    _user="${SLIM_USER%%:*}"
-    _user=$(slim_resolve_user "$_user")
-    su "$_user" -c "$_cmd"
-    poweroff -f
-}
-
-# === Execute the configured command ===
-# If slim.cmd=<base64> is on the kernel cmdline, decode and exec it.
-# Otherwise, exec /slim/exec (generated from the image's CMD/ENTRYPOINT).
-# shellcheck disable=SC2013 # word-splitting is intentional: cmdline tokens are space-separated
-for tok in $(cat /proc/cmdline 2>/dev/null); do
-    case "$tok" in
-        slim.cmd=*)
-            v=${tok#slim.cmd=}
-            decoded=$(b64dec "$v") || echo "base64 decode failed"
-            if [ -n "$decoded" ]; then
-                slim_exec "$decoded"
-            fi
-            ;;
-    esac
-done
-slim_exec "/slim/exec"

src/virtiofs.rs

@@ -0,0 +1,222 @@
+//! Virtiofs: spawn and manage one virtiofsd daemon per `--mount` share.
+//!
+//! virtiofsd is a vhost-user device backend that runs in a separate
+//! process from QEMU and serves files from a host directory over the
+//! FUSE protocol via a UNIX socket.  boco spawns one daemon per share,
+//! hands the socket path to QEMU via `-chardev socket,path=...`, and
+//! cleans up when the VM exits.
+//!
+//! To serve files owned by arbitrary guest UIDs (e.g. when an image
+//! uses the `USER` directive) without running as root, virtiofsd is
+//! launched under `podman unshare` in a user namespace with
+//! `--sandbox chroot`.  This reuses rootless podman's existing
+//! subuid/subgid mapping: guest UID 0 maps to the invoking user, and
+//! other guest UIDs map into the subuid range, so each guest UID
+//! becomes a distinct host UID.
+//!
+//! Lifecycle: boco spawns all virtiofsd children before launching QEMU,
+//! waits for each listening socket to appear, then drops the
+//! [`VirtiofsShare`] guard after QEMU returns — `Drop` sends SIGTERM,
+//! waits briefly, escalates to SIGKILL, and `TempDir` removes the
+//! socket files.  On signals that kill boco directly (SIGINT/SIGTERM
+//! without a handler), the guard does not run and the daemons linger;
+//! callers like `example/test.sh` should `pkill virtiofsd` in cleanup.
+
+use std::path::{Path, PathBuf};
+use std::process::{Child, Command, Stdio};
+use std::thread;
+use std::time::{Duration, Instant};
+
+use anyhow::{Context, Result, anyhow, bail};
+use tempfile::TempDir;
+
+use crate::command::cmd;
+
+/// Locations searched for the virtiofsd binary, in order.
+const VIRTIOFSD_CANDIDATES: &[&str] = &[
+    "/usr/libexec/virtiofsd",
+    "/usr/lib/qemu/virtiofsd",
+    "/usr/lib/virtiofsd",
+];
+
+/// How long to wait for virtiofsd to create its listening socket.
+const SOCKET_READY_TIMEOUT: Duration = Duration::from_secs(5);
+const SOCKET_POLL_INTERVAL: Duration = Duration::from_millis(10);
+/// Grace period after SIGTERM before escalating to SIGKILL.
+const STOP_GRACE: Duration = Duration::from_secs(2);
+
+/// A running virtiofsd daemon plus the resources it owns.
+///
+/// `Drop` stops the daemon and removes its socket.  The `TempDir`
+/// holding the socket is owned by the share so its path stays valid
+/// for the lifetime of `socket`.
+pub struct VirtiofsShare {
+    /// Tag advertised to the guest (e.g. `boco0`).
+    pub tag: String,
+    /// Guest-side mount destination, as supplied by the user (may be
+    /// relative to the image's WORKINGDIR; resolved by the guest init).
+    pub guest_path: String,
+    socket: PathBuf,
+    child: Option<Child>,
+    _dir: TempDir,
+}
+
+/// Locate the virtiofsd binary.
+///
+/// Searches well-known distro paths first, then falls back to `$PATH`
+/// lookup via `command -v`.
+pub fn find_virtiofsd() -> Result<PathBuf> {
+    for path in VIRTIOFSD_CANDIDATES {
+        if Path::new(path).exists() {
+            return Ok(PathBuf::from(path));
+        }
+    }
+    let which = cmd(&["sh", "-c", "command -v virtiofsd"])?;
+    let trimmed = which.trim();
+    if trimmed.is_empty() {
+        bail!(
+            "virtiofsd not found. Install it (often shipped with qemu, or \
+             build https://gitlab.com/virtio-fs/virtiofsd) and ensure it is \
+             in PATH or at one of: {}",
+            VIRTIOFSD_CANDIDATES.join(", ")
+        );
+    }
+    Ok(PathBuf::from(trimmed))
+}
+
+fn resolve_host_dir(host: &str) -> Result<PathBuf> {
+    let path = Path::new(host)
+        .canonicalize()
+        .with_context(|| format!("Cannot resolve mount path '{host}'"))?;
+    if !path.is_dir() {
+        bail!("Mount source '{}' is not a directory", path.display());
+    }
+    Ok(path)
+}
+
+/// Wait for `socket` to appear as a filesystem entry.
+fn wait_for_socket(socket: &Path) -> Result<()> {
+    let start = Instant::now();
+    while start.elapsed() < SOCKET_READY_TIMEOUT {
+        if socket.exists() {
+            return Ok(());
+        }
+        thread::sleep(SOCKET_POLL_INTERVAL);
+    }
+    bail!(
+        "virtiofsd did not create socket at {} within {:.1}s",
+        socket.display(),
+        SOCKET_READY_TIMEOUT.as_secs_f64()
+    );
+}
+
+/// Spawn one virtiofsd daemon for a share.
+///
+/// `index` becomes the suffix of the tag (`boco{index}`) and is used to
+/// derive a unique socket filename.
+pub fn spawn_share(index: usize, host_path: &str, guest_path: String) -> Result<VirtiofsShare> {
+    let virtiofsd = find_virtiofsd()?;
+    let canonical = resolve_host_dir(host_path)?;
+    let host_str = canonical
+        .to_str()
+        .context("Mount source path is not valid UTF-8")?;
+    let virtiofsd_str = virtiofsd
+        .to_str()
+        .context("virtiofsd path is not valid UTF-8")?;
+
+    let dir = TempDir::new()?;
+    let tag = format!("boco{index}"); // short tag to fit on cmdline
+    let socket = dir.path().join(format!("{tag}.sock"));
+
+    // TODO: Make rootless mode optional
+
+    // Run virtiofsd under `podman unshare` so it can serve files owned
+    // by arbitrary guest UIDs without root, mapping them into the host
+    // user's subuid range.  `--sandbox chroot` is required because the
+    // outer `podman unshare` already entered a user namespace (nested
+    // user namespaces are not allowed); chroot still prevents symlink
+    // escapes from the shared dir.
+    let mut command = Command::new("podman");
+    command
+        .args(["unshare", "--", virtiofsd_str, "--socket-path"])
+        .arg(&socket)
+        .args([
+            "--shared-dir",
+            host_str,
+            "--tag",
+            &tag,
+            "--sandbox",
+            "chroot",
+            "--cache",
+            "auto",
+        ])
+        .stdin(Stdio::null())
+        .stdout(Stdio::null())
+        .stderr(Stdio::inherit());
+
+    println!("{command:?}");
+    let child = command
+        .spawn()
+        .with_context(|| format!("Failed to spawn virtiofsd: {command:?}"))?;
+
+    wait_for_socket(&socket)
+        .with_context(|| anyhow!("virtiofsd for share '{tag}' failed to become ready"))?;
+
+    Ok(VirtiofsShare {
+        tag,
+        guest_path,
+        socket,
+        child: Some(child),
+        _dir: dir,
+    })
+}
+
+impl VirtiofsShare {
+    /// Socket path to pass to QEMU's `-chardev socket,path=...`.
+    pub fn socket_path(&self) -> &Path {
+        &self.socket
+    }
+
+    /// Send SIGTERM, wait briefly, then SIGKILL if still alive.
+    fn stop(&mut self) {
+        let Some(child) = self.child.as_mut() else {
+            return;
+        };
+
+        // SIGTERM first for a graceful exit.
+        let pid = child.id();
+        let _ = Command::new("kill")
+            .arg("-TERM")
+            .arg(pid.to_string())
+            .stdin(Stdio::null())
+            .stdout(Stdio::null())
+            .stderr(Stdio::null())
+            .status();
+
+        let start = Instant::now();
+        while start.elapsed() < STOP_GRACE {
+            match child.try_wait() {
+                Ok(Some(_)) => {
+                    self.child = None;
+                    return;
+                }
+                Ok(None) => thread::sleep(Duration::from_millis(50)),
+                Err(_) => break,
+            }
+        }
+
+        // Escalate to SIGKILL.
+        if let Some(child) = self.child.as_mut() {
+            let _ = child.kill();
+            let _ = child.wait();
+        }
+        self.child = None;
+    }
+}
+
+impl Drop for VirtiofsShare {
+    fn drop(&mut self) {
+        self.stop();
+        // TempDir drop removes the socket file and its parent dir.
+    }
+}